Ray and I are part of a small group interested in verified coding agents. He has a habit of tracking down the reference behind the reference. Last week his post Zig Creator Calls Spade a Spade, Anthropic Blows Smoke hit the top of Hacker News. 1,546 points. 784 comments. It stayed there.
The controversy
Anthropic's Claude helped Bun rewrite itself from Zig to Rust — presented as a technical decision driven by memory safety. Andrew Kelley, Zig's creator, pointed out the codebase was a mess due to engineering decisions, including overusing AI agents. Zig was blamed for problems it didn't cause.
AI is not enough
Ray's core insight: every element of the story that Bun and Anthropic present as evidence that AI replaces engineers actually demonstrates the opposite.
The million-line agentic PR? Required human review and cleanup the AI couldn't do. The automated language migration? Left semantic gaps only a human could close. The speed of the rewrite? Achieved by a team working 90-hour weeks, not by AI magic. The case study Anthropic chose to prove software engineering is becoming obsolete is contradicted by every layer of its own story.
The thesis isn't "AI is bad." It's "AI is not enough." The AI was a tool. The humans were the engineers. The story Anthropic needs to tell — to justify a $132 billion valuation — is disproven by the very evidence they selected.
The reference density
The post cites 38 references: DARPA TRACTOR evaluation reports, TigerBeetle's TigerStyle methodology, Buddhist right-speech frameworks, Ed Zitron's financial analysis, Hillel Wayne's empirical SE talk. Each is woven into the argument, not dropped for credibility. It's the kind of writing that takes a week to produce and a career of reading to make possible.
Go read it
Ray's post is about Zig and Rust and Anthropic and Bun. The lesson is about something bigger: how to think clearly when the loudest voices in the room have $132 billion worth of reasons to be heard.
Zig Creator Calls Spade a Spade, Anthropic Blows Smoke by Ray Myers. Hacker News discussion (1,546 points, 784 comments).